Digital image‐color conversion between different illuminants by color‐constancy actuation in a color‐vision model based on the OSA‐UCS system

Abstract

In digital image capture, the camera signals produced by the D65 illuminant, once translated into tristimulus values of the CIE 1931 standard colorimetric observer (assuming the Maxwell‐Ives‐Luther criterion is satisfied), are considered good to produce accurate color rendering. An image obtained under any illuminant other than D65 does not appear realistic and the tristimulus values of the camera must be transformed into the corresponding ones produced by the D65 illuminant. This transformation must satisfy color constancy. In this work, the transformation is obtained by a color‐vision model based on the Optical Society of America‐Uniform Color Scales system [Color Res Appl 2005; 30: 31–41] and is represented by a matrix dependent on the adaptation illuminant. This matrix is obtained by minimizing the distance between the pairs of the uniform scale chromatic responses related to the tristimulus values of the 99 different color samples of the SG Gretag‐Macbeth ColorChecker measured under a pair of different illuminants, one of which is the D65. Then any picture captured under a given light source can be translated into the picture of the same scene under the D65 illuminant. Metameric reason allows only approximate solutions. The transformations from Daylight and Planckian illuminants to the D65 illuminant have a very regular dependence on the color temperature, that appears to be the typical parameter for the color conversion. © 2012 Wiley Periodicals, Inc. Col Res Appl, 38, 412–422, 2013

Color constancy from invariant wavelength ratios: An algorithm to simplify the chromatic adaptation model

“…The model's inherent simplicity entirely avoids the veritable minefield of conventional chromatic adaptation models including such issues as the relative advantages/disadvantages of different chromatic-adaptation primaries and their effect on color gamut, for example the spectral sharpening or purple problem, and of course the considerable and ever-changing math complexities of Von Kries (e.g., CIE-CAT02) or two-matrix chromatic adaptation transforms. [14][15][16][17] In their place, admittedly, the present model introduces the (relatively minor) complexity of equivalent wavelength for nonspectral hues, which diminishes after initial use.…”

Chromatic adaptation by illuminant matrix products: An alternative to sharpened von kries primaries

“…In searching for the best x ‐space representations of A , B , C , Oleari started with the cardinal dichromat confusion vectors P , D , and T which are traditional Von Kries primaries. After this, H matrices were bootstrapped to different illuminant–observer combinations by minimizing over a particular set of reflectances i the root‐mean‐square value of the quantities Δ E i , where …”

“…The derivation of matrices T and H by the above method employs a foundation in color constancy (illuminant invariance), rather than in the results of asymmetric‐matching experiments. The method was found to be at least as effective as CAT02 in predicting the change of tristimulus values due to illuminant change—quite a different goal from comparison with corresponding color data from human subjects. Nonetheless, the two‐matrix framework may be able to be repurposed for fitting such data.…”

“…That separation is also essential when one considers a new kind of solution to the sharpening dilemma: allow the matrices H 1 and H 2 a completely general dependence on their respective illuminants, rather than requiring them to be diagonal in the same basis (as is true for the Von Kries rule). This has been done by Oleari et al 8,9 Without envisioning the adaptation in two stages, it is impossible even to imagine how a chromaticadaptation matrix between two illuminants could be selectable by illuminant, because it would have to be selectable by two illuminants-old and new! But by thinking about the adaptation process as having two segments, one can associate a single illuminant with each segment, and hence each segment has its own matrix.…”

“…In searching for the best x-space representations of A, B, C, Oleari started with the cardinal dichromat confusion vectors P, D, and T which are traditional Von Kries primaries. After this, 8,9 H matrices were bootstrapped to different illuminant-observer combinations by minimizing over a particular set of reflectances i the root-mean-square value of the quantities DE i , where…”